inspect
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from external web pages to identify and resolve elements.
- Ingestion points: Browser DOM data is ingested through
mcp__chrome-devtools__take_snapshotand specific element properties (text, IDs, classes, data attributes) are retrieved viamcp__chrome-devtools__evaluate_scriptinSKILL.md. - Boundary markers: None present. The agent processes text and attributes directly as identifiers for resolving element locations.
- Capability inventory: The skill uses
mcp__chrome-devtools__evaluate_scriptfor browser interaction andRead/Grepfor local source file analysis, with the potential for subsequentEditoperations on the code. - Sanitization: There is no evidence of sanitization or escaping for element text or attributes before they are used in search queries or presented to the user.
- [DYNAMIC_EXECUTION]: The skill utilizes
mcp__chrome-devtools__evaluate_scriptto execute a JavaScript closure at runtime within the browser context to extract element metadata (tags, IDs, classes, and data attributes). While the script itself follows a static template, the execution is dynamic and depends on the state of the web page being inspected.
Audit Metadata